Spillover Effects on Caregivers’ and Family Members’ Utility: A Systematic Review of the Literature

We included peer-reviewed articles published in English that reported a preference-based measure of caregiver or family member utility or disutility, including caregiver-focused measures for which population tariffs exist (i.e., CarerQol [20] and CES [21]), from the inception of each database through 3 April 2018. We included articles that reported on multiple patient diseases and/or using multiple preference-based methods/instruments, but excluded articles that reported only the EQ-VAS or a visual analog scale measure unless the scores were transformed into utilities using a known algorithm [22].

We defined family member as anyone identified as having a familial relationship to the patient regardless of distance (e.g., cousins would meet our inclusion criterion). We assumed that all family members classified in articles as caregivers were such; we did not impose any criteria on this role. We included articles reporting on ‘informal caregivers’ unless they were described as exclusively non-familial, such as neighbors, church members, and the like, but excluded paid caregivers.

We included articles reporting on all patient diseases and conditions, including those that specified no disease, meaning they included caregivers regardless of the patient’s disease. We also included articles reporting on patient health states, defined as a distinct phase of a disease or condition (such as chemotherapy or hospitalization); a disease was defined as a diagnosed condition. We excluded disease transmission among family members from our definition of spillover effects. We included death as a health state when it was directly related to a disease or condition, such as maternal mortality, but we did not specifically search for bereavement. We imposed no age limit on patients or caregivers/family members.

We included articles reporting on studies specifically designed to measure spillover utility, those measuring caregiver/family member utility among other outcomes, and caregiver or patient interventions that included utility as an outcome. We excluded reviews, reports, study protocols, commentaries, editorials, and conference papers, as well as articles that reported what appeared to be invalid utilities, such as scores > 1.0 or those described as ‘WTP utilities’.

After excluding duplicates, two authors (EW and LJ) independently screened titles and abstracts; conflicts were resolved by consensus. We repeated this process with the full-text articles remaining after screening. We recorded the reason for each exclusion using Covidence systematic review software (Veritas Health Innovation, Melbourne, VIC, Australia).

We extracted data that would allow a reader to identify potentially useful values for an analysis: patients’ disease/condition; patients’ age (adult/child/either); valuation measure used (EQ-5D, standard gamble, etc.); sample source (e.g., medical centers, patient association, population); country of sample; affected person’s role (i.e., family member/family caregiver/informal caregiver); caregiver/family member age (mean or other summary measure); sample size; utility (mean or median); if relevant to study design: comparison group source, sample size, and utility (mean or median); and if relevant, the reporting of utilities by strata, and other notes. We created a table entry for each patient disease/condition for which a relevant utility was reported in an article; articles that reported utility for more than one patient disease/condition were included in an entry for each. We included multiple utilities measured using different methods (e.g., HUI2 and HUI3) or applying different valuation weights for the same measure (e.g., Canadian and US weights for SF-6D) in one entry. If both caregiver/family member utilities and spillover disutilities were reported, we included each. If utilities were reported for the same condition/disease for multiple countries, we reported the one with the largest country-specific sample size. For all other instances of multiple utilities reported for the same disease/condition, we included those we deemed most salient to most readers and noted the availability of others in the ‘notes’ comment. We included the scores/values as reported by authors, but performed no manipulations or calculations on reported data. We grouped the entries into three categories: (1) spillover utilities or disutilities; (2) caregiver and/or family member utility reported with a matched or population comparison group; and (3) caregiver and/or family member utility reported alone.

‘Spillover’ results from caregiving, simply caring about others, or a combination of the two. What constitutes caregiving for one person may be ordinary behavior for another—the distinction, for example, between ‘regular’ parenting and caregiving for sick children [33]. Caregiving is often shared among family members, it sometimes vacillates between family and paid caregivers (such as when patients cycle between home and hospital), and it changes in nature and intensity over time [34]. Caregiving can provide sustenance to family members who might otherwise feel a lack of control or disengaged from an ill relative, moderating the otherwise burden of care [5]. While the literature tends to focus on primary caregivers, other family members both provide care and experience spillover effects of illness [5, 35]. The inconsistent description of caregiving and family involvement in the literature injects variability into both the estimation of spillover QALYs and how to interpret the policy implications of family-based CEAs.

Health utility scores can capture the health-related spillover effects of caregivers and family members if the health-specific spillover effects are isolated; meaning, the utility associated with solely the caring for or caring about component of having an ill relative. These utility scores can be used to calculate QALYs for CEAs. Care-related measures, such as the CarerQOL, albeit preference-based, include non-health domains in addition to health, and as such are incompatible with CEAs. However, utility scores that reflect only the change in health-related quality of life (HRQOL) associated with spillover are available for just a small set of conditions, therefore other measures of caregiver/family member effects may yield values suitable for use in CEAs when certain assumptions hold.

Most articles in this review fall into the category of conventionally-defined health utility scores: the utility of a caregiver or non-caregiving family member of an ill relative. These scores may include the impact of spillover but also the underlying health of the individual. Elderly caregivers, for example, are likely to have chronic health conditions simultaneous with their caregiving responsibilities, therefore their utility scores will reflect a combination of both effects. In some of these articles, utilities are reported for a matched sample, such as non-caregivers or family members of healthy individuals, or from general population norms, allowing for the calculation of a ‘spillover utility’. In the majority of articles reviewed, however, a comparison group utility is not reported, but an analysis could use an appropriate population norm to derive a spillover utility, carefully considered to match the demographics of the caregiver sample. The underlying assumption in this literature is that spillover effects are additive, yet this has not been empirically demonstrated; interaction effects have been hypothesized [30]. While a growing supply of spillover utility data is available in the literature, much of it requires assumptions such as this to be able to use these utilities in a CEA. A further challenge is that the current literature rarely distinguishes between caring for and caring about effects, which may be difficult to disentangle.

Caregiver-focused ‘utility equivalents’ are preference-based but are distinct from QALYs. These measures—the CarerQol and CES—include a different and more comprehensive set of dimensions than are typically included in QALY-based measures (e.g., fulfillment, financial problems, relational problems). While they may accurately capture caregiver-relevant dimensions [36], their valuations are based on a care-related quality-of-life scale so cannot be used to estimate QALYs, and therefore can neither be combined with patient QALYs in CEAs nor compared with CEAs based on QALYs [20]. They are of particular value, however, for comparatively evaluating caregiver interventions. An early prototype of a caregiver-specific measure that was QALY-based but focused on caregiver-relevant dimensions—the Caregiver Quality of Life Index (CGLI)—was largely supplanted by these instruments [37].

Incorporation of spillover effects into CEAs faces significant methodological questions, including concerns regarding prioritizing caregiver health over patient health, equity in decision making, and ‘double counting’ of benefits. At the most basic level, spillover and patient QALYs can be summed to arrive at total QALYs, and this is currently the most common approach observed in practice. Some have proposed a weighting factor be applied to adjust for the relative importance of spillover effects compared with changes in health for the primary patient [14]. In the extreme, including spillover QALYs could tilt decisions toward benefiting caregivers/family members over the patient, although this is not the intended purpose of considering these effects [38]. Whose QALYs to include in spillover is also an unanswered question as evidence suggests effects extend beyond the primary caregiver [39] and perceptions of relevance (i.e. ‘closeness’) vary across individuals [40]. The articles in this review provide the data to inform family-based CEAs but do not inform the questions underlying appropriate incorporation approaches.

Equity issues are significant in including spillover effects in CEAs. If spillover is included, cost-effectiveness ratios for interventions targeting diseases/conditions that require caretaking and/or negatively affect family members could be favored over diseases/conditions that do not. Interventions that affect oftentimes isolated patients, such as homeless individuals, could be undervalued relative to those that affect more connected individuals, such as children. At the same time, not including spillover could result in policy decisions that disregard the interests of caregivers and families [38, 41]. While it might be sensible to consider the effect of pediatric conditions on parents as well as the child, as successful treatment confers benefits on both, in cases such as dementia, patients’ and families’ interests may at times be at odds: successful treatment may prolong the patient’s life and extend the caretaking burden for the family. Including spillover QALYs in CEAs is both a methodological decision, including the ‘what’ and ‘how’ aspects, and a normative decision, including the ‘if’ and ‘when’ aspects. Whether it is normatively justifiable to include spillover in economic evaluation recognizing the potential reallocation of resources that may ensue is as yet unresolved [38, 41]. It is clear, however, that including spillover necessitates the accurate and normative definition of ‘who qualifies’, as who is or is not included can have an impact on the results, regardless of how those results are used in policy.

Another concern raised with regard to including spillover in CEAs is double counting: spillover effects may already be implicitly included in utilities. Patients’ anxiety or depression, for example, may be a function of their condition’s effect on their family, rather than or in addition to its effect on themselves, indicating that family spillover may be at least in part reflected in patient utilities. Moreover, it may be difficult for caregivers or family members to disentangle their health from their ill relative’s, therefore their ‘spillover’ utility may include more than the effect on themselves individually. Although reasonable concerns, difficulty in measurement ought not preclude the incorporation of an endorsed component of effects. Advances in measurements, most of which use direct utility elicitation methods, have been made to attempt to accurately capture spillover independently, although with unconfirmed success [27, 28, 30]. Judicious use of sensitivity analysis may be a reasonable approach for the time being to minimize the effect of potential measurement error on CEA results.

Considerations should be noted regarding our review. While our search is comprehensive as of our end date, articles including utilities for spillover effects are being published with increasing frequency [19], and will soon render our catalog incomplete. We are in the process of developing an online, open-access repository of spillover effect utilities, which will be updated regularly as a public resource. Moreover, our catalog excludes the gray literature or unpublished sources. Unpublished utilities that are in the pipeline, via conference presentations and abstracts, will likely find their way into the published literature in the future and will be incorporated into successive versions of the catalog. Publication bias is not a concern for this review because our results are descriptive and not intended for inference. We limited the data included in our tables to ensure accessibility for readers—essentially a size that was viewable on a standard size page or computer screen—therefore details that are important to some may have been omitted. Finally, we made subjective judgments about the relative salience of utilities in articles reporting multiples, but describe others in the ‘notes’ section of the tables.